In a conventional image processing device, as disclosed in Japanese Unexamined Patent Publication no. 125857/1996 (Tokukaihei 8-125857, published on May 17, 1996), first and second characteristic parameters are found and inputted to a determination circuit using a nerve circuit network so as to perform area determination (area separation) of a target pixel. Here, the nerve circuit network is a non-linear type and has been learned in advance. Besides, the non-linear type means that inputs of first and second characteristic parameters are respectively converted to coordinates on a vertical axis and a horizontal axis, and a separating state is shown on the coordinates.
When performing area separation using the above non- linear separating method, it is necessary to widely memorize coordinates. These coordinates are called a lookup table, which is adopted for converting an output based on an input axis. Therefore, such a lookup table uses a memory for storing data. Further, the conventional arrangement has required considerably large memory.